Statistical Programs |
College of Agriculture | University of Idaho |
Seminar Announcement |
"Applied Statistics in Agriculture" |
Regularization in Skewed Binary Classification Presented By |
Dr. Stephen Lee |
Division of Statistics University of Idaho |
Tuesday, April 15 3:30 P. M. Ag. Science 62 |
For some binary classification problems, for example the
presence or absence of a rare disease, the frequencies of
the two classes are highly unbalanced. The ratio between
the classes could be as high as 1:100 or even more. This
may cause serious problems to many classification models
since not enough information is available for the rare
class. We approach the problem by repeating the rare cases
with or without adding noise to the training data, and
comparing the predictive classification performance on an
unaffected (no repeats, no perturbations) test data. The
models considered include: classic discriminant methods,
logistic regression, nearest neighbor methods, classification
trees, and neural networks applied to several real world and
simulated data sets. Encouraging results, in terms of
increased ROC area, were obtained in some models.
All interested faculty, staff, and graduate students are invited to attend. |
Return to Seminar Listing. |